WaveRider LMS4000, Руководство пользователя

Страница: 60 / 282

3 Detailed Description
42 APCD-LM043-4.0
Transmit Queue Limits
CCU transmit buffer space is a limited resource shared between the EUMs. If more traffic is
received at the CCU for transmission to an EUM than can actually be transmitted to it, that
EUM might eventually use up all available CCU buffer space, effectively starving all other
users. Therefore, the number of packets in each EUM's transmit queue is intentionally limited.
Packets arriving beyond this limit are discarded, resulting in retransmission of TCP/IP packets
by the host computer and TCP/IP adjusts by slowing down. The EUM transmit queue length
limit, which is never less than the lower bound given in the GOS parameter set, is dynamic and
based on total queue occupancy.
EUM transmit queue length limit determines the optimal TCP receive window size (the
maximum allowed number of outstanding unacknowledged bytes) used by the host
application. Some Internet Speed Boost programs intended for DOCSIS or ADSL
connections, simply increase the receive window size to very large values. This increase
results in very long queues at the CCU, more discarded packets, increased retransmissions,
and reduced throughput. To maximize throughput, WaveRider recommends setting the
receive window size of these applications to between 18000 bytes (~12 packets) and 24000
bytes (~16 packets).
TIP: Utilities are commercially available for optimizing the TCP
receive window size in the end-user’s computer, through
manipulation of the Windows registry.
Polling MAC Statistics
A wide range of Polling MAC statistics are recorded by the CCU and EUM. These statistics are
very useful, particularly during installation and as an aid to troubleshooting. A complete list of
statistics provided by entering the <stats mac> command through the CLI can be found in
Monitoring the Network on page 127.
Performance Modelling
The performance of packet radio systems like the LMS4000 900MHz Radio Network cannot
easily be derived from analytic calculations. However, using computer simulations that are
designed to accurately reflect the system implementation, and user and network traffic
distributions, it is possible to produce statistical representations of LMS4000 system
performance.
WaveRider has developed a model that simulates LMS4000 system processes, tasks,
protocols, propagation delays, and queue sizes. The model can simulate systems with large
numbers of EUMs and wide ranges of user traffic. The inputs to the model include
• number and geographical distribution (distance from CCU) of EUMs,
• user traffic statistics, and
• RF link-quality distributions.
These inputs are based on WaveRider’s experience with actual customer installations. The
outputs of the model are statistical representations of system performance.